In the breeding of grain sorghum (Sorghum bicolour L. Moench) towards improved nutrition and agronomic performance, new methodologies are required to increase genetic diversity and lower the inputs required to track and screen breeding populations. Near-infrared calibration models were developed by partial least squares (PLS) and test-set validation on 364 sorghum samples to predict crude protein and moisture content on whole-grain and milled flour samples. Models using milled flour spectra were more accurately predictive than those from whole grain spectra for all constituents (eg. Protein: R2 = 0.986 on flour vs R2 = 0.962 on whole grain). Discriminant calibrations were established to classify grain colour using partial least squares discriminant analysis (PLS-DA) based upon CIE L*a*b* reference values and visual ranking. Preliminary calibrations were developed for quantities of 18 amino acids, fat and apparent metabolisable energy (AME) on 40 samples using cross-validation, highlighting potential for reliable calibration for these parameters in sorghum. An investigation into the potential of 12C6+ heavy-ion beam mutagenesis of sorghum seed was undertaken by treatment at RIKEN Accelerator Research Facility (Saitama, Japan) and subsequent breeding at Ukulinga research farm and analysis at the Department of Plant Pathology, University of KwaZulu-Natal, Pietermaritzburg, South Africa. Dosage rates of 75, 100 and 150 Gy were compared in seven sorghum varieties to establish optimal dose treatments as determined by germination and survival rates, visible morphological changes and field data over two seasons of field trials. Crude protein variation within the M2 generation was analysed to compare dose rate effects. The need for higher dose rates was indicated by few quantified differences between treatments and control although good correlations between protein deviation and treatment dose rate were elucidated. Differences in varietal response suggest a need to optimize dose rate for specific varieties in future endeavours. In addition, all mutagenized populations were screened for crude protein content using near-infrared spectroscopy (NIRS). Significant differences in protein levels and standard deviations were observed between treated self-pollinated M2 generations and untreated control populations. Individual plants displaying significantly different protein levels were isolated. / Thesis (M.Sc.)-University of KwaZulu-Natal, Pietermaritzburg, 2009.